An Incremental Dynamic Time Warping for person re-identification

Abstract

This paper presents principles and techniques of a human gesture recognition algorithm for person identification which identifies personal gait patterns recorded with a 3D depth sensing camera, in this case the Microsoft Kinect® version 2. The recorded images are analyzed against a dataset of gait gestures derived from a sample of 37 people. We compared two algorithms for analyzing movement trajectories; Sparse code and Incremental Dynamic Time Warping (IDTW). Experimental results show that the methods have an encouraging performance. When comparing the accuracy of algorithms, IDTW gave better recognition results than the Sparse code method.